©
Magmatic Processes: Physicochemical Principles
The Geochemical Society, Special Publication No. I, 1987
Editor B. O. Mysen
Dry peralkaline felsic liquids and carbon dioxide
flux through the Kenya rift zone
D. K. BAILEY
Department of Geology, The University, Whiteknights,
Reading RG6 2AB, Berkshire, U. K.
and
R. MACDONALD
Department of Environmental Sciences, University of Lancaster, Lancaster LA! 4YQ, U. K.
Preamble- When Hat Yoder was in London to receive the Wollaston Medal of the Geological Society the chance
arose for him to see the Michaelangelo Tondo in the Royal Academy. His reaction, naturally, was to marvel that
such beauty could be wrought from stone. Along with other participants at the meeting we pay tribute to Hat, who
in his own way has continually given petrologists new perceptions of rocks.
Abstract-Around
Lake Naivasha, at the topographic culmination of the Kenya rift, a wide range
of peralkaline felsic magmas erupted in the Holocene. Glassy samples are abundant. Volcanological
and petrographic evidence indicates that these glasses reflect characteristically low concentrations of
H20 in the melts, and this relationship is confirmed by experimental and chemical information.
Distinct magma types, trachytes, comendites and pantellerites, erupted almost exclusively from
different centres, even though these are sufficiently close for intercalation of the erupted products
from adjacent volcanoes. This individuality is maintained in chemical variation diagrams where the
products from different centres form separate groups, with no gradational compositions. In some
diagrams smaller clusters appear within the main groups. These distinctive patterns indicate that
there is no continuous evolution of the melts below the rift, and that each centre represents the
outpouring of a different melt. Batches within the same centre could also represent different episodes
of melt generation.
Generation of these melts in a regional context can best be explained by melting of different sources
during the influx of a CO2-rich fluid. Trachytes, pantellerites and comendites could then be the
products of a flux/melting cycle as it climbs progressively through mantle, mafic deep crust, and
sialic crust. The high concentrations of sodium and iron in the melts call for enhancement of these
elements in the source rocks: this could be achieved by regional metasomatism resulting from prolonged
activity through the rift segment over the past 23 Ma. Incompatible trace element variations cannot
be attributed to solid-melt interactions but could be imparted by the fluid during melting.
INTRODUCTION
Aphyric volcanic glasses provide the closest
samples to the composition of peralkaline melts,
where it is clear that crystallisation, and indeed devitrification or hydration, leads notably to alkali and
halogen losses (NOBLEet al., 1967; MACDONALD
and BAILEY,1973). Glasses may also provide extra
insight into the gases associated with strongly alkaline volcanism. This information is of value because the only direct, high temperature volcanic
gas collection available is from the melilite nephelinite volcano of Mount Nyiragongo (CHAIGNEAUet al., 1960): no direct, high temperature collections have been made of an oversaturated peralkaline volcano. The aim here is to examine and
explore the potential information content of a range
of oversaturated volcanic glasses from the central
part of Kenya (Gregory) rift valley. All the samples
are fresh, unaltered glasses; most are obsidians in
which phenocrysts are absent or less than 2 percent
in the mode; vesicles, if present, are widely sepa-
FORA VARIETYof reasons alkaline rocks have always held a fascination for igneous petrologists, but
in recent years have assumed extra significance as
it has emerged that their richness in incompatible
elements and radio-isotopes are signalling special
conditions of source, or melt-forming processes, or
both. In most alkaline provinces peralkalinity has
developed in varying degrees. Paradoxically, peralkalinity is strongest when magma penetrates continental lithosphere, where the crustal rocks are
typically peraluminous. Such magmatism is especially evident in stable plate interiors where igneous
activity of any kind must call for special explanation. Even where eruption is localised by an evident
split in the lithosphere, there is still no ready explanation of why, for instance, continental fissuring
may give vent to tholeiite floods in one region or
epoch, and alkaline magmatism in another.
9!
92
D. K. Bailey and R. Macdonald
(2) Samples are from all geographic and tectonic
settings.
(3) Wide variations in vesicularity, and in phenocryst types and contents, are represented.
INTRINSIC H20 CONTENTS OF
(4) Eruptive mode varies widely. Samples come
PERALKALINE GLASSES
from flow margins (top and bottom); flow interiors;
With one exception all peralkaline glass analyses angular blocks in agglomerate; angular fragments
in the compilation by MACDONALDand BAILEY in pumice deposits; fiamme from welded ash flows;
(1973) have H20+ contents of 0.67 weight percent
splatter; and agglutinated spatter.
or less. Many more analyses (published and unSignificantly, tephra fragments ranging from
published) have since been added (BAILEY,1978;
small chips in pumice deposits to large blocks in
1980) and it is clear that the vast bulk of the H20+
agglomerates are all low in H20+; there is no evivalues are around the limit of detection (~0.1
dence of high speed ejectamenta that could repreweight percent): hence, most glasses are virtually
sent an H20-rich melt.
anhydrous. These data are in complete accord with
the case put forward earlier by NICHOLLSand CARExperimental crystallisation and
MICHAEL(1969) that oversaturated peralkaline liqcomparative petrography
uids are essentially dry. Other studies (see EWART,
1979) seem to indicate consistently low/02 and high
In marked contrast with calc-alkaline rhyolites,
temperatures in calc-alkaline rhyolites and dacites, which fail to crystallise in the laboratory when dry,
which would be in keeping with the generally low pantellerite glasses start to crystallise in as little as
H20+ values in the glasses. Recently, however, thirty minutes at appropriate temperatures without
TAYLORet al. (1983) have reported H20 contents
addition of water. Experimental study of the melting
up to 3.1 weight percent in fresh calc-alkaline rhy- and crystallisation of peralkaline glasses under a
olite tephra, leading to the suggestion that these are range of conditions is, therefore, possible (COOPER,
the general levels in such melts, and that glasses 1975; BAILEYand COOPER,1978) with results as
with low H20 are degassed. They propose that the
shown in Figure 1. Details of the experimental conmain degassing of the melt takes place during
ditions and the phases developed were given in
BAILEYand COOPER(1978), which also contains a
foaming within the vent, so that later flows and
domes with low H20 represent extrusion of the col- discussion of the consequences of the contrasting
lapsed foam. They report that there appear to be behaviour of wet and dry pantellerite melts. Releno differences in the compositions of the calc-al- vant points from this discussion are summarised
kaline glasses except for H20+ and deuterium con- below.
tents. In view of the manifest compositional changes
(1) The exsolution of vapour under near-liquiin peralkaline glasses during devitrification and hydus conditions in the dry experiments (Figure 1)
dration, and during high temperature fusion exindicates that some natural volatiles were quenched
periments (HAMPTONand BAILEY, 1985), any
in at partial pressures at least as high as 0.3 kbar.
massive loss of several percent H20 from a high
A synthetic rhyolite melt would contain 2.5 weight
temperature peralkaline melt would render the
percent of H20 at an equivalent pressure (LUTH,
consequent anhydrous glass suspect as a sample of
1976). Addition of H20 to the natural pantellerite
the natural melt composition. It is essential, thereprofoundly changes the melting behaviour indicatfore, to examine the evidence with this in mind
ing that it is capable of dissolving a similar quantity
before considering the chemical variations among
of H20 at 0.3 kbar. As H20 is one of the most
the glasses from the Kenya rift province.
soluble gasesin silicate melts it hardly seems credible
that a pantellerite melt could lose 2-3 weight perGeneral relationships and field evidence
cent H20 whilst effectively retaining other gases at
The case presented here is part of a continuing
0.3 kbar solubility levels. Hence, although there
study of alkaline glasses, encompassing the widest
must have been some degassing from some samples
possible range of bulk compositions and samples
(in decompression) this gas was not simply H20.
from all regions and all different modes of eruption.
(2) Crystallisation of an H20-rich pantellerite
The following list outlines the range and scope.
results in acmitic pyroxenes and potassic feldspars
quite different from those observed as natural phe(1) Bulk compositions range from nephelinite
nocrysts, and in the wrong sequence.
through phonolite to comendite, i.e., the maximum
(3) Kenyan obsidians with modal phenocryst
range of silica values.
rated. Pumice samples are excluded because of their
susceptibility to post-eruptive hydration and alteration.
Peralkaline
felsic liquids
93
Am
OIL
Ac\L
+
V
P
kb
700
800
900
1000
T °C
FIG. 1. Comparison of hydrous and anhydrous crystallisation of a pantellerite obsidian (sample
KE12: Eburru, Kenya). Hydrous boundaries, heavy lines: anhydrous boundaries, light. The dry subsolidus region is shaded. V = fluid, L = liquid; Q = quartz; Af = alkali feldspar; AfK = K-rich alkali
feldspar; Af, = alkali feldspar critical line (LuTH, 1976); Cpx = sodic hedenbergite; Am = arfvedsonitic
amphibole; Ac = acmitic pyroxene; S = dry solidus. The points shown at 1 kbar are the range of
liquidus minima (quartz + feldspar + liquid) in related synthetic systems at PH20 = 1 kbar (CARMICHAEL and MACKENZIE, 1963; THOMPSON and MACKENZIE, 1967). Taken from BAILEY and
COOPER (1978).
contents greater than 5% are rare: most have less
than 2% and, hence, have a near-liquidus aspect.
This characteristic, combined with low H20+ contents, effectively rules out an origin from a magma
at, or below, the H20-saturated liquidus in Figure
1. Firstly, the large differences in temperature between the wet and dry systems, require that subvolcanic degassing of any wet pantellerite magma
should have been accompanied by extensive crystallisation. Secondly, the negative slopes of the phase
field boundaries in the wet system would imply that
any wet pantellerite magma at depth could get to
the surface only as a pyroclastic eruption composed
of H20-saturated glass, or as degassed crystalline
lava.
If pantelleritic melts ever were wet at depth then
some special conditions of eruption would be called
for to account for the dry, near-liquidus aspect of
the glasses.Suppose the original melt had previously
contained as much as 3 weight percent H20 (e.g.,
approximately saturated at 2 km depth) and had
somehow lost H20 at a lower pressure. To achieve
a dry near-liquidus condition its required starting
temperature would be at least 50°C above the wet
liquidus. Alternatively, it might have risen towards
the surface from a point on a 3 weight percent H20
liquidus, requiring a starting pressure of greater than
5 kbar. To make the transition from wet to dry
melt therefore would seem to require an unusual
eruption or thermal history. Non-systematic (vagarious) variation within the sample population
would be expected in H20 contents, oxidation
states, alkalies and halogens. Furthermore, when
crystals are present, at least some should carry a
record of this eruption history. In fact, they are remarkably consistent in composition (NICHOLLS
and
CARMICHAEL,
1969; SUTHERLAND,1974) even to
the extent of showing scarcely any sign of zoning
in the feldspars, the dominant phenocryst phase.
Peralkaline glasses are still puzzling in the light
of their laboratory behaviour. Dry pantellerite held
94
D. K. Bailey and R. Macdonald
just above its solidus temperature for about an hour
starts to crystallise; if taken to the liquidus temperature it bubbles. But both these states seem to have
been largely missed out in the cooling of natural
glass samples. In order to preserve such a glass it
must somehow pass through the vesiculation and
crystallisation temperature range (below about
700°C at low pressure) in a short time. Whereas
this cooling history can be envisaged for particular
instances such as flow margins, it is less easy in
other cases. Glass clasts in pumice deposits are typically non-vesicular, non-porphyritic and angular.
Hence, they were "through" this temperature range
before incorporation in the pumice eruption. Our
experiments with gas extraction indicate that foaming is a near-liquidus phenomenon (HAMPTONand
BAILEY,1984). It is difficult, therefore, to see how
the obsidian clasts could have come from the part
of the melt that was foaming. If they are chilled
melt from the border zone of the foaming mass
they should presumably record the highest H20
contents available for peralkaline liquids.
Glass fiamme in welded ash flows also pose
problems if, as suggested by TAYLORet al. (1983),
they "remain at magmatic temperatures for weeks
or months". According to our experiments pantellerite fiamme could not survive as glass under such
conditions. Almost wholly glassy flows and domes
are also extruded from peralkaline silicic volcanoes
and must pose similar problems. There is a need
for some means of super-cooling peralkaline melts
prior to eruption such that the viscosity is sufficiently high to inhibit crystallisation and bubbling,
but flow is permitted.
obsidians show no correlation between H20 and Cl
(Figure 2). In fact, no correlation between H20 and
any other variable has been found in these rocks,
certainly not with alkalies and halogens where it
might be most expected. Present evidence offers no
case for massive sub-volcanic loss of water-rich vapour from peralkaline melts. The low and variable
H20 levels in the glasses may be essentially "accidental", i.e., unrelated to the melt generation process. On this basis it is appropriate to treat the glasses
as melt samples, in an attempt to unravel their
chemical relationships and try to discern their petrogenesis.
PERALKALINE VOLCANOES IN THE
CENTRAL KENYA RIFT
Volcanism has been active along the Kenya rift
for the past 23 Ma, ranging from carbonatites and
nephelinites to rhyolites. Felsic rocks constitute
about half the total erupted volume, about 150,000
knr' (WILLIAMS,1982). The bulk of the TertiaryRecent felsic rocks are peralkaline making it the
most voluminous and variegated peralkaline province in the world. Holocene activity in the central
part of the rift, around its topographic culmination
in the vicinity of Lake Naivasha, has been largely
expressed by felsiccentral volcanoes on the rift floor.
Figure 3 shows the largely monotonic magma compositions emitted from the main centres. Eruptives
from adjacent centres may interfinger around the
margins, but are always petrographically distinct.
The general geologyaround Lake Naivasha has been
described by THOMPSONand DODSON(1963), and
there have been individual descriptions of Suswa
(JOHNSON,1969); Longonot (SCOTT, 1980); and
Chemical variations and H20+ levels
If peralkaline magmas have previously contained
several percent H20, they must clearly have suffered
massive vapour loss prior to glass formation. This
must leave its mark on the melt composition, as
other volatile species should be partitioned into the
vapour. In high-temperature, gas extractions from
glasses in the laboratory, H20 is always accompanied by other species, either in the form of gases,
or precipitated as sublimates. Most typically, the
high-temperature sublimates include alkali halides
(HAMPTONand BAILEY,1985). These products
mirror those observed at high temperature fumao
roles on volcanoes. Similarly UNNIand SCHILLING
1000
5000
CI ppm
(1978) have recorded that concentrations ofCI and
H20 in submarine basalts correlate with ocean
FIG. 2. Variation ofCI with H20 in oversaturated perdepth, and are dependent on degree of vesiculation
alkaline glasses. Diamonds indicate samples from Panteland loss of vapour on eruption at different pressures. leria, all other samples from the Kenya rift (for details see
In contrast to these observations, peralkaline silicic Figure 4).
Peralkaline
....
. .' . . .
:.:-::: N' :'.<-:
'
felsic liquids
95
(MCCALL, 1967; MACDONALD et al., 1970), and
the East African rift valley has so far provided the
only samples of this melt. Trachyte glasses having
clear chemical affinities with comendites are absent.
The extent and persistence of peralkaline activity
through the East African lithosphere make it imperative to look for the causes, and the wide range
of geology and sample compositions offer hope for
eventual solutions to the questions of petrogenesis.
It is appropriate first to look at chemical variations
in the glasses with changing levels of peralkalinity.
Major element variations with peralkalinity
FIG. 3. Sketch map showing the chief Holocene volcanic
centres in the Lake Naivasha section of the Kenya rift.
Rift shoulders indicated by stipple.
south west Naivasha (MACDONALD et al., 1986). In
addition to the felsic volcanoes, there is a small basalt field forming the Badlands, north east of Eburru
(MCCALL, 1957). Small basalt cones pepper the
fringes of the comendite field of south west Naivasha, and mixed basalt/felsic lavas have been
erupted from Longonot (SCOTT and BAILEY, 1984)
and Eburru. North of the Badlands the next major
central volcano is Menengai, erupting only trachyte
(LEAT, 1984).
Lake Naivasha is located on the crest ofthe classic
continental rift valley first described by GREGORY
(1894) and its volcanism has a special place in igneous petrogenesis. Many of the lavas used by
BOWEN (1937) to illustrate his concept of Petrogeny's Residua System (nepheline-kalsilite-silica)
were from this region. The Holocene volcanism
around Lake Naivasha is an almost perfect case for
a petrographic province, fulfilling all the criteria laid
down by WILCOX (1979).
The Lake Naivasha region has the added advantage that glassy samples, covering most ofthe range
of oversaturated
peralkaline
compositions
are
abundant. The first pantelleritic trachyte glasses
were recognised
at Menengai
and Longonot
Some distinctions are seen immediately. For instance, the comendites from south west Naivasha
separate from the pantellerites and the pantelleritic
trachytes in terms of Si02, Ti02, FeO, MnO and
total iron (Figures 4 and 6). They also appear as a
distinct group on most other variation diagrams.
Like all other continental comendites, they appear
to fall close to the quartz-feldspar cotectic, as extended into the peralkaline region (BAILEY and
MACDONALD, 1970, Figure 5; MACDONALD et al.,
1986, Figure 7), with the implication that the source
contained free quartz. This characteristic, together
with many other features of their major element
composition would be most obviously satisfied by
a crustal or sialic source (BAILEYand MACDONALD,
1970) and more detailed studies on trace elements
and isotopes are consistent with this interpretation
(MACDONALD et al., 1986; DAVIES and MACDONALD, 1986).
Figure 4 also reveals separate clustering of trachytes and pantellerites, which persists in Al203
(Figure 5). As might be anticipated, there is a general
negative correlation between alumina and peralkalinity, but the comendites, pantellerites and trachytes form separate groups. This separation is emphasized when molecular Ah03 is plotted; the
comendites and pantellerites appear to define one
trend, while the trachytes form parallel trends at
higher levels. This feature might be taken to reflect
different series of liquids with various levels of silica,
but it is noteworthy that trachyte liquids show no
direct connection with rhyolites.
Just as it might be anticipated that Al203 would
correlate negatively with peralkalinity, so a positive
correlation might be expected for the alkalies (Figure
5). In comendites and pantellerites, K20 is almost
constant around 4.4 weight percent, but there are
signs of slight negative distributions, which are emphasised when the trachytes are taken into account.
The general fall in K20 levels with increasing peralkalinity may be linked in some way to the "or-
D. K. Bailey and R. Macdonald
96
75
•
.,
. ...
++
1·0
&.
0
0
"i
0""
00
Ti02
Si02
%
j,
%
..
'.,
0·5
I
0
0,
I
-l
I
~
0
0)
'"
00
0
'"
:DO
...
0
'.,
,
0
0
"
.
",0
D
D
'0.
1
65
..
'.
+
+
0·1
60
1·0
2·0
PERALK
1·0
PERALK
2·0
0'4
TRACHYTES
o
o
0·3
MnO
00
%
o
0·2
'"
0(1)
•
EBURRU
•
LONGONOT
D
MENENGAI
PANTELLERITES
o
EBURRU
o
PANTELLERIA
0
COMENDITES
0·1
+
NAIVASHA
"
HIPPO
"MIXED"
1·0
PERALK
POINT
LAVAS
2·0
FIG. 4. Variations ofSi02, Ti02, and MnO with peralkalinity [Mol (Na20 + K20/AI203)]
in glassy
samples from the Naivasha region. Plus signs, comendites from the main Naivasha field; cross, indicates
tight group of comendites from the Hippo Point area on Lake Naivasha; open circles, pantellerites,
closed circles, trachytic pantellerites from Eburru; filled squares, trachytes from Longonot; open
squares, trachytes from Menengai; symbols with diagonal line indicate mixed lavas (possibly accumulitic
in the case of Menengai).
00
0
Peralkaline felsic liquids
"
.
'
97
I.
5-0
15·0
+
..
+
.="
++
..
+
++
10·0
OJ
+
3·0
o
o
o
00
o
o
cc
1·0
1'0
2·0
PERALK
12
·0,--------------
__ -,
90
•
• 0
11·0
10·0
2·0
PERALK
i
8·0
",
..
9·0
all
C
AI203
Mol.%
o
'.
8·0
'"
o
:
e
c
++
7·0
:t.
+
",
6·0
++
6·0
-s
5·0i--.~r_~_r_.--r_.__r_.--r_._~
1·0
PERALK
°cc
I
+
I
+
5·0
2·0
1·0
PERALK
2·0
FIG. 5. Variations of Ah03 (weight percent and mol percent), Na20 and K20 with peralkalinity.
Mol percent Ah03 can be used as an indication of variation in potential feldspar in the samples.
Symbols as in Figure 4.
thoclase effect" (BAILEY and SCHAIRER, 1964)
whereby alkali feldspars tend to be richer in K20
than their coexisting melts. It is not easy, however,
to reconcile the relatively constant K20 in the comendites and the pantellerites with this effectif indeed
they represent series of liquids controlled by closedsystem fractionation.
An even more striking pattern emerges for Na20
(Figure 5). Trachytes from Longonot and Eburru
fall on quite separate positive trends, which them-
D. K. Bailey and R. Macdonald
98
...
'"
o
7·0
10·0-1
00
~
.. .
9·0
o
=Jp
o
6·0
0
o
o
7·0
5·0
FeO
1·".,'.,
FeOT
%
%
OCD
00
80
CD
'0.
6·0
4·0
5·0
3·0
.. ..
2,0-
+
4·0
..
+
+
+
3·0
'·0
2·0
2·0
PERALK
1
'·0
1
1
2·0
PERALK
10·0
3·0
• •••
• •a
9'0
o
8·0
.
o
e
o
. .,
•
2,0-
o
c
!lib
o
.,
7·0
,
FeOT
0
0
%
6·0
I-
o
o
,~o
.,
,
'.
D,
'0
e
.0
+
"
r
5·0
+
4·0
+
+
3·0
1·0
2·0
1·0
PERALK
2·0
x
5·0
8·0
9·0
FIG. 6. Variations of soda and iron for Kenyan glasses. Symbols as in Figure 4.
selves are distinct from the distributions of pantellerites and comendites. This emphasizes the distinctions seen in Ah03 and is mirrored by variation
in other major elements, particularly FeO and total
iron (Figure 6). Discrete populations of liquids are
identified, some of which have been erupted even
from the same volcano (Eburru). Within any series
there is a general positive correlation between Na20
and total iron, which may reflect some internal
consistency in melt generation within that series,
Peralkaline
but the separation of the series, especially trachytes
from rhyolites, would seem to require generation
at least in distinct batches. Increasing soda correlates
with iron content, and within each group with peralkalinity.
Between trachytes and pantellerites,
however, there is a broad negative distribution
which seems to defy explanation in terms of interactions between melts and their observed phenocryst phases.
Where these rocks are not aphyric, the ubiquitous
phenocryst is alkali feldspar, and the bulk compositions of the melts require that this phase must
dominate in any crystallisation sequence. Separation of potassic feldspar from the trachytic liquids
must lead to increasing sodicity beyond that actually
observed in the pantellerites (MACDONALD et al.,
1970). Therefore, if these liquids were linked in a
crystallisation sequence, a sodic phase was removed
from the magma system as crystallisation
progressed: no appropriate composition appears as a
phenocryst in the trachytes.
The distribution of soda and iron (Figures 5 and
6), confirm the earlier suspicion that there is no
chemical continuum between trachytes and pantellerites, and that they developed under different
conditions. Pantellerite may represent a quartzfeldspar cotectic composition for a specific range of
source rock compositions and pressures. At higher
pressures (and in different source compositions?)
the cotectic melt is closer to trachyte, as with increasing pressure in the "granite system" (TUTTLE
and BOWEN, 1958). The special requirement in the
"pantellerite system" would be that there is a phase
boundary (marking the appearance of a phase rich
in soda and iron) imposing a limit on the peralkalinity of the liquid that can be reached at any
given pressure (see BAILEY, 1974).
A plot ofFe203 vs peralkalinity (Figure 6) reveals
a steady progression through the comendites that
could reflect an increasing contribution of acmite
to the melts. There is no such regular variation of
ferric iron and peralkalinity in the trachytes or pantellerites, and the latter appear to mark a "ceiling"
in the possible levels of Fe203 in peralkaline melts.
As the trachytes also have more iron than comendites or pantellerites, it is not possible to appeal
simply to an acmitic component to account for the
Fe203 variations in this part of the array. Melting
of a source containing a combination
of phases
similar to the observed phenocrysts (feldspar, fayalite, sodic hedenbergite, aenigmatite, and oxide)
could produce the trachytes. As discussed later,
pantellerite melt generation may need to involve
an arfvedsonitic amphibole in addition.
Formation
of these peralkaline trachyte and
felsic liquids
99
pantellerite liquids seems to require melting of appropriate source compositions. This is indicated by
the marked composition gaps, which have persisted
despite intensive sampling, and must depict a real
distribution of natural melts. The gaps must militate
also against generation of trachytes and pantellerites
by progressive melting of a single source. The separation of the trachytes from Longonot, Menengai
and Eburru in various diagrams also renders unlikely the possibility of fractional melting at separate
trachyte and rhyolite invariant points, for which
there is no experimental evidence in any case. The
separation of trachytic and pantelleritic liquids is
further emphasized by the fact that in many diagrams the variations within the groups run across
any expected trend from trachyte to pantellerite
(Figure 4, 5 and 6). Further discussion of the questions of melt generation is best deferred until trace
element variations have been examined.
Trace element versus major element functions
When variation in incompatible trace element
concentrations is compared with peralkalinity the
patterns are different (Figure 7). The main Naivasha
comendites show a positive variation, but no other
simple relationship emerges between trace elements
and peralkalinity. Patterns in the trachytes are not
well defined but some signs of positive trends may
be detected within groups. The pantellerites, however, are scattered, with the highest levels of inc ompatible elements near the lowest levels of peralkalinity. Essentially, the trachytes, pantellerites and
comendites form separate populations
(c.f., the
major element data), but distributions within groups
are hinting at further differences between them. One
major difference becomes clear when traces are
compared with soda and iron. For the comendites
there is a positive correlation of trace elements with
Na20, but in the pantellerites the broad distribution
is negative (Figure 8). Because FeO and FeOT correlate positively with Na20, their relationships with
trace elements are similar. What also emerges in
Figure 8 is distinct clustering of different parts of
the pantellerite population.
Variations between trace elements
Trace element correlations are much stronger.
The strong correlation of Zr with Rb and F that
was first noted for Eburru pantellerites (BAILEYand
MACDONALD, 1975) appears in all groups of liquids
even though they may separate from each other in
the diagrams (Figure 9). This relationship has been
found also in sample suites from the Azores and
from Pantelleria (BAILEY, unpubl.).
100
D. K. Bailey and R. Macdonald
o
400
D
o
500
o
o
...
.
..
300
y
e
o
400
..
Nb
ppm
ppm
o
+
300
;'00-
••
100
.
200
D.
'.
"~D..
100
T
1·0
2·0
PERALK
1'0
2·0
PERALK
..
8000,-----------------
-,
o
++
CD
....
400
.•.
o
o.
F
ppm
300
5000
Rb
ppm
o
o
08
200
CD
"'
•• all
'"
100
'.,
1·0
PERALK
I000-f·'
r
1·0
2·0
FIG. 7. Variations with peralkalinity
glasses. Symbols as in Figure 4.
i
I
I
[
I
Iii
PERALK
for a selection of trace elements,
Coherence also exists for Nb and Y through all
the groups, and it had been noted previously
(BAILEY and MACDONALD, 1975) that in the Eburru
obsidians these two elements were more strongly
correlated with Cl, as distinct from the strong co-
I
I
I
2·0
Y, Nb, Rb and F, in Kenyan
variance in F, Zr and Rb. When all groups of glasses
are compared, the pattern of Cl distribution is less
clear cut, and there is a strong suggestion in Figure
9 of decoupling of the CI distribution from variations in the other trace elements in the higher con-
I
Peralkaline felsic liquids
9·0
I
D
T···
'".
8·0
NozO
.
%
'.
7°l
.
g-
"
e
e
e
-'1'
g
.
. .
5·0
5000
1000
Cl ppm
9'O~--------------r---------------'
8·0
NoZO
.
~oJ.
Comendites, pantellerites and trachytes form
distinct populations, and for some elements, subgroups of these populations emerge. Three recurring
sub-groups in the trachytes relate directly to the
source volcanoes, Menengai, Longonot and Eburru.
Eburru trachytes are distinct from Eburru pantellerites in most plots. Within individual groups there
is a general trend of increasing soda and iron with
peralkalinity, with the pantellerite groups lying on
the extension of the comendite trend (Figure 6).
Superimposed on this, however, there is a broad
negative distribution from trachyte to pantellerite.
Furthermore, although incompatible elements show
positive covariance within the groups, they correlate
negatively with soda and iron in the pantellerites,
in contrast with trachytes and comendites. Viewed
as a whole, it is clear that more than one process
has been at work in generating these liquids. Soda
(and iron) and peralkalinity seemingly can vary independently of incompatible trace elements.
'.
.
. .-
.
.
s oj
I
I
1060
5000
F ppm
9·0
I
D
,f':
e 0
I
No2O
%
.l".
·D.
7
g'
".
"1
",
. .
centration ranges. There appears to be a "ceiling"
for Cl in Kenyan glasses, which also shows up dramatically in the plot of'Na-O vs. Cl (Figure 8), where
the negative distribution of the pantellerites and the
positive distribution in the comendites appear to
curve towards a common maximum. This contrasts
with the Na20 vs F diagram in which the more
linear distributions might be pointing to a maximum outside the observed range. This limitation
in the chlorine appears to be a regional effect because liquids from Pantelleria achieve much higher
levels ofCl, describing a linear array with no obvious
"ceiling" (Figures 8 and 9). The Kenyan rift liquids
point to a limit in available Cl, but not F, in the
source. The simple conclusion reached previously
for the Eburru pantellerites that F, Zr and Rb
formed one chemically coherent group, whilst Nb,
Y and Cl formed another, therefore needs modification. A more reasonable conclusion would be that
Zr and Rb cohere very strongly with F, whereas Nb
and Y partly cohere with F, with part of their distribution connected with availability of Cl.
Summary of the chemical distributions
m,
D,
.
G·O_]
101
THE SIGNIFICANCE OF SEPARATE
POPULATIONS OF FELSIC LIQUIDS
.. ..
:
5·0
4·6
1000
2000
3000
Zr ppm
FIG. 8. Variations between different trace elements and
Na20 for peralkaline glasses from Kenya. Diamonds
indicate samples from Pantelleria. Other symbols as in
Figure 4.
The chemical variation diagrams emphasise the
individuality of different volcanic centres by showing that each has erupted distinct and separate felsic
magmas. Furthermore, there are contemporaneous
basalts and hawaiites, but no extruded rocks with
silica percentages between 54 and 61 weight percent
that are not mixed lavas.
Different volcanoes have, therefore, erupted
products that are not only chemically distinct, but
102
D. K. Bailey and R. Macdonald
8000
I
between which there are no bridging compositions.
These distinct felsic magmas are all products of a
major igneous cycle working through this segment
oflithosphere. They are coeval and share some distinctive chemical features, especially peralkalinity.
What is their relationship? If a basaltic parent were
to be invoked, then each felsic batch has to evolve
quite separately to produce the batches under the
different volcanoes. In all cases the parent and the
process must lead to strong peralkalinity and great
enrichment in incompatibles. But if each magma
type represents merely a stage in a common evolutionary process why should anyone stage be restricted to one volcano? The most ready alternative
is that the different volcanoes do not represent various degrees of melt differentiation but are different
:
.. ..
*
I
F
ppm
5000
0
I
0
o
=. .
0
0
0
0
0
g,.
D: •
0
".
1000-1··...
melts.
1000
2000
3000
Zr ppm
e
'0
0
..~
I
F
:
ppm
5000
I
0
o~o
0
"'.-
..
~
0
..
'0,
'0.
1000-1::
1000
5000
CI ppm
.
400
.
=' •
1
300
y
'::j
:
....
0
=8.
. .
00
100~
",
Closed-system magma evolution can be discounted even for single complexes, such as Eburru,
where Rb varies as if it were an incompatible element (BAILEYand MACDONALD,1975). This relationship cannot be reconciled with generation of
the liquids solely by crystal-liquid interactions because the distribution coefficient for Rb in alkali
feldspar is greater than 0.3. The behaviour of Rb
in a series ofliquids in equilibrium with alkali feldspar, based on measured distribution coefficients
for Rb and Zr and the Rayleigh fractionation law
is shown for the Eburru sequence in Figure 10. Also
shown are the actual and calculated distributions
for the Naivasha comendites and pantellerite glasses
from Pantelleria. In all three cases the predicted
curves depart from the actual distributions, and it
must be assumed that in each series some factor
other than evolution by feldspar-liquid interaction
is necessary to give the observed distribution pattern
for rubidium. As halogen contents typically rise
above one weight percent in the more siliceous liquids, no mineral seen in the rocks could provide
the source for incompatible elements. One possibility is that a halogen-rich fluid was contributing
to the source region from the onset of melting.
Alternatively the high levels of incompatible elements might perhaps be attributed to late-stage
processes after the melt had filled a shallow magma
chamber. Some limited variation is possible by
crystal fractionation, as inferred for Longonot
(SCOTT,1980) but the Rb factor rules this process
out for Eburru (BAILEYand MACDONALD,1975)
and for Naivasha comendites (MACDONALD
et al.,
1986). Diffusion, or "liquid state" processes, may
account for part of the variation, as suggested from
·D.
~
1000
5000
Cl ppm
FIG. 9. Variations between trace elements for peralkaline
glasses. Diamonds indicate samples from Pantelleria. Other
symbols as in Figure 4.
Peralkaline felsic liquids
the Kenya rift zone (BAILEY, 1978, 1980). The main
points, with some additions, may be briefly summarised as follows.
++
I
400
103
++
/
/
300
Rb
/
ppm
200
100
1000
Zr ppm
2000
3000
FIG. 10. Variation between Rb and Zr in peralkaline
glasses. For three groups (comendites, and pantellerites
from Eburru and from Pantelleria) the calculated alkali
feldspar fractionation curves are shown by short dashed
lines with arrows. The distribution coefficient for Rb (D
= 0.3) is a mean from values given in BAILEY and MAcDONALD(1975) and additional determinations, in which
the distribution ofZr (D = 0.046) was also measured (using
separated feldspars and glass matrices: Eburru samples;
BAILEY, unpubl.). Symbols as in Figure 4.
the evidence of zoned ash flows on Menengai (LEAT
et al., 1984), but if such a process has given rise to
zoned magma chambers, the products lie within
the range for the magma type, which itself still remains tightly constrained
in anyone
complex.
Concentration
of incompatible elements solely by
diffusion also runs into difficulty with low and nonsystematic variations of H20, and the apparent
ceiling on Cl levels in the Kenya province. Difficulties are further compounded by the fact that the
distribution pattern between Na, Fe, and incompatible elements is reversed in pantellerites compared with trachytes and comendites.
High level, magma chamber processes offer no
solutions to the fundamental geological questions
of how these distinct felsic magma types have been
produced, how they have preserved their individuality, and what regional conditions have led to all
the individual centres developing such striking
richness in volatile and incompatible elements. An
explanation that aims to encompass these factors
must consider melt sources and the possible influence of fluids in magma generation.
CARBON DIOXIDE FLUX
THROUGH THE RIFT
In previous papers attention has been drawn to
the evidence for massive emission of CO2 through
(1) Igneous activity was initiated by nephelinite/
carbonatite volcanism, which has continued intermittently for the past 23 Ma. An example of the
early activity is the nephelinite volcano of Tinderet,
with associated Miocene carbonatite ashes (DEANS
and ROBERTS, 1984) 50 km to the west of the trachyte volcano ofMenengai. The active carbonatite
volcano of Oldonyo Lengai at the southern end of
the rift is also part of regional igneous activity that
includes felsic volcanism. In nephelinite volcanic
gases, CO2 dominates over H20; in nephelinite
glasses from Nyiragongo, CO2 is about ten times as
abundant as H20; and carbonatite lavas from 01donyo Lengai are anhydrous (but rich in F and Cl)
(BAILEY, 1980).
(2) At the present time, all along the rift there is
escape of CO2, which is sufficiently abundant to be
exploited in commercial wells at several places. This
CO2 flux has led to the suggestion by Kenyan geologists that present day geothermal fields along the
rift are the result of heating of the ground water by
hot juvenile CO2 (WALSH, 1969).
(3) Due to the low solubility of CO2 in salic
magmas, evidence of the former presence of CO2
in Kenyan peralkaline rocks is bound to be scarce,
nevertheless, high-temperature
carbonates are recorded in small pockets in some of the trachyte
lavas on Longonot (SCOTT, 1982). A mixed carbonate/glass ash flow and carbonate lapillae tuffs
have been discovered on Suswa (SKILLING and
MACDONALD,personal communication,
1986). The
main commercial CO2 well also happens to be on
the edge of the rift just to the east of Suswa.
The Kenya rift is fizzing with CO2 and has been
for at least the past 23 million years. This narrow
gash in the lithosphere acts as a release channel for
volatiles escaping from a large mantle volume below, analogous to the function of a pie funnel
(BAILEY, 1980). The general thermal, chemical, and
geodynamic consequences of such fluid focussing
have been explored elsewhere (BAILEY, 1983).
SYNTHESIS
In the absence of any evidence for continuous
melt evolution, the objective procedure would be
to treat each magma batch as an entity, or the product of a separate event. Any links within or between
the major groups, trachytes, pantellerites,
and
comendites, must then be attributable to common
features in the melt generating regime. From this
starting point, the following deductions are possible
for the melting system:
104
D. K. Bailey and R. Macdonald
(1) Comendites show regular increase in Fe203
with peralkalinity, consistent with control by melting of an acmitic pyroxene in the source. A sodic
pyroxene could exert some influence on the trachyte
variations. Any such pyroxenes could be formed by
metasomatism prior to melting.
(2) By contrast, pantellerites show no regular
variation in Fe203, even though Na20 and FeOT
rise steadily through the range. This relationship
could describe an increasing contribution from a
stable amphibole (fluor-arfvedsonite?) at increased
melt levels. Such a melting pattern would give rise
to higher sodium and iron with greater melt volumes, consistent with the negative correlation between Na20 and incompatible elements in the pantellerites. It would imply also that any fluorine contribution from the amphibole was insufficient to
produce a noticeable effect on the distribution pattern of F in the pantellerites as a whole.
(3) Halogens and other incompatible elements
are mainly contributed from a fluid entering the
melt zone. The anticipated general result would be
higher concentrations in earlier, smaller melt volumes, but some variations in the melt samples could
be due to varying amounts of fluid flux through the
melt zone (see 4.d).
(4) Part of the fluid has very low solubility in
siliceous liquids, with the effect that it is not wholly
consumed in the initial melt. Consequently:
(a) Elements are partitioned between melt and
fluid.
(b) Partitioning is dependent on melt composition, e.g., trachyte or rhyolite.
(c) Partitioning is dependent on pressuretemperature conditions, and source mineralogy.
(d) Because part of the fluid is insoluble in the
melt there can be fluid flux through the melt region.
Thus each melt may be buffered for incompatible
elements by the fluid.
(5) The fluid is essentially anhydrous and CO2rich. It also carries other volatile elements such as
alkalies (which can be contributed to the melt) but
its most striking effect is on the incompatible element budget of the melt. Partitioning works both
ways and the fluid scavenges alkaline earths (notably
Ba and Sr) from the more siliceous melts. This
would account for the extremely low Ba and Sr
concentrations in these melts, which defy explanation by closed-system crystal fractionation or
partial melting hypotheses.
(6) The three main magma types are determined
by three types of metasomatised source rocks:
(a) comendite from metasomatised sialic
rocks;
(b) pantellerite from metasomatised basic
rocks;
(c) trachyte from metasomatised mantle containing felsic minerals (BAILEY,1986).
(7) Different magma types show major differences in their chemical patterns. These might simply
reflect fundamental differences in the fluid source,
but there is a strong possibility that some of the
differences may be a consequence of interaction between the fluid and rocks in its path before entering
the melt zone. For instance, the higher Rb in comendites may be related to earlier scavenging of
this element as the fluid moved through crustal
rocks before entering the melt zone. On the other
hand, the generally lower levels of incompatible
elements in the trachytic melts may be more a
function of variations in the partitioning between
the CO2-rich fluid and a less siliceous melt at higher
pressures.
(8) Individual magma types erupt from distinct
centres depending on the source level in the lithosphere. Trachyte centres tap a melt source in metasomatised mantle. As the trachyte source is exhausted, and the melting cycle climbs higher in the
lithosphere, pantellerite can become the dominant
eruptive. Comendite represents the ultimate stage
of the rifting/melting process, when the melting cycle finally impinges on metasomatised sialic crust.
Where an igneous cycle is well established, a range
of sources, from mantle to crust, may be providing
melts to a series of contemporaneous volcanoes.
This scenario is consistent with the late development
of the Lake Naivasha felsic magma centres, almost
symmetrically disposed around the topographic
culmination of the Gregory Rift.
Acknowledgments-We
thank our friends and colleagues
in the Department
of Geology, University of Reading,
whose skills and cooperation have been vital in this study.
Fieldwork in Kenya was made possible by grants from the
Natural Environment
Research Council.
Tables of analysis are available from the first author.
REFERENCES
BAILEY D. K. (1974) Experimental petrology relating to
oversaturated peralkaline volcanics: a review. Bull. Volcanol. Special Vol. 38,635-652.
BAILEY D. K. (1978) Continental rifting and mantle degassing. In Petrology and Geochemistry of Continental
Rifts, (eds. E. -R. NEUMANN and I. B. RAMBERG), pp.
1-13 Reidel.
BAILEY D. K. (1980) Volcanism, earth degassing and replenished lithosphere mantle. Philos. Trans. Roy. Soc.
London A297, 309-322.
Peralkaline
BAILEY D. K. (1983) The chemical and thermal evolution
of rifts. Tectonophys. 94, 585-597.
BAILEY D. K. (1986) Fluids, melts, flowage and styles of
eruption in alkaline ultramafic magmatism. Trans. Geol.
Soc. S.A. (in press).
BAILEY D. K. and SCHAIRER J. F. (1964) Feldspar-liquid
equilibria in peralkaline liquids-the
orthoclase effect.
Amer. J. Sci. 262, 1198-1206.
BAILEY D. K. and MACDONALD R. (1970) Petrochemical
variations among mildly peralkaline (comendite) obsidians from the oceans and continents. Contrib. Mineral. Petrol. 28, 340-351.
BAILEY D. K. and MACDONALD R. (1975) Fluorine and
chlorine in peralkaline liquids and the need for magma
generation in an open system. Mineral. Mag. 40,405-
14.
BAILEY D. K. and COOPER J. P. (1978) Comparison of
the crystallisation of pantelleritic obsidian under hydrous
and anhydrous conditions. In Progress in Experimental
Petrology, NERC Publications Series D 11, 230-233
Eaton Press.
BOWEN N. L. (1937) Recent high-temperature
research
on silicates and its significance in igneous geology. Amer.
J. Sci. 33, 1-21.
CARMICHAEL I. S. E. and MACKENZIE W. S. (1963) Feldspar-liquid equilibria in pantellerites: an experimental
study. Amer. J. Sci. 261, 382-396.
CHAIGNEAU M., TAZIEFF H. and FABRE R. (1960) Composition des gaz volcaniques du lac de lave permanent
du Nyiragongo. Acad. Sci., Paris, C. R. 250,2482-2485.
COOPER J. C. (1975) Experimental melting and crystallization of oversaturated peralkaline glassy rocks. Ph.D.
Thesis, University of Reading.
DAVIES G. R. and MACDONALD R. (1986) Crustal influences in the petrogenesis of the Naivasha basalt-rhyolite
complex: combined trace elements and Sr-Nd-Pb isotope constraints. (Submitted to J. Petrol.).
DEANS T. and ROBERTS B. (1984) Carbonatite tuffs and
lava clasts of the Tinderet foothills, western Kenya: a
study of calcified natrocarbonatites. J. Geol. Soc. London
141, 3, 563-580.
EWART A. (1979) A review of the mineralogy and chemistry of Tertiary-Recent
dacitic, latitic, rhyolitic, and
related salic volcanic rocks. In Trondhjemites, Dacites,
and Related Rocks (ed. F. BARKER), pp. 13-121 Elsevier.
GREGORY J. W. (1894) Contributions to the physical geography of British East Africa. Geogr. J. 4, 290.
HAMPTON C. M. and BAILEY D. K. (1984) Gas extraction
experiments on volcanic glasses. J. Non-Cryst. Solids
67, 147-168.
HAMPTON C. M. and BAILEY D. K. (1985) Sublimates
obtained during fusion of volcanic glasses. J. Volcanol.
Geotherm. Res. 25, 145-155.
JOHNSON R. W. (1969) Volcanic geology ofMt. Suswa,
Kenya. Phi/os. Trans. Roy. Soc. London 265A, 383-
412.
LEAT P. T. (1984) Geological evolution of the trachytic
caldera volcano Menengai, Kenya Rift Valley. J. Geol.
Soc. London 141, 6, 1057-1069.
LEAT, P. T., MACDONALD R. and SMITH R. L. (1984)
Geochemical evolution of the Menengai Caldera Volcano, Kenya. J. Geophys. Res. 89, 8571-8592.
LUTH W. C. (1976) Granitic rocks. In The Evolution of
felsic liquids
105
the Crystalline Rocks, (eds. D. K. BAILEY and R. MACDONALD), pp. 335-417 Academic Press.
MCCALL G. J. H. (1957) Geology and groundwater conditions in the Nakuru area. Tech. Rept. 3, Hydraulic
Branch, M.O.W. Nairobi.
McCALL G. J. H. (1967) Geology of the Nakuru-Thomson's Falls-Lake
Hannington area. Geol. Surv. Kenya,
Report No. 78.
MACDONALD R. and BAILEY D. K. (1973) The chemistry
of the peralkaline oversaturated obsidians. u.s. Geol.
Surv. Prof Paper 440N-I.
MACDONALD R., BAILEY D. K. and SUTHERLAND D. S.
(1970) Oversaturated peralkaline glassy trachytes from
Kenya. J. Petrol. 11, 507-17.
MACDONALD R., BLISS C. M., LEAT P. T., BAILEY D. K.
and SMITH R. L. (1986) Geochemical evolution of the
Naivasha peralkaline rhyolitic complex, Kenya: I. Mineralogy and Geochemistry. (submitted to J. Geol.).
NICHOLLS J. and CARMICHAEL J. S. E. (1969) Peralkaline
acid liquids: a petrological study. Contrib. Mineral. Petrol. 20, 268-294.
NOBLE D. c., SMITH V. C. and PECK L. C. (1967) Loss
of halogens from crystallized and glassy silicic volcanic
rocks. Geochim. Cosmochim. Acta 31, 215-223.
SCOTT S. C. (1980) The geology of Longonot volcano,
Central Kenya: a question of volumes. Philos. Trans.
Roy. Soc. London 296, 437-465.
SCOTT S. C. (1982) Evidence from Longonot volcano,
Central Kenya, lending further support to the argument
for a coexisting CO2 rich vapour in peralkaline magma.
Geol. Mag. 119,215-217.
SCOTT S. C. and BAILEY D. K. (1984) Coeruption of contrasting magmas and temporal variations in magma
chemistry at Longonot Volcano, Central Kenya. Bull.
Volcanol. 47, 849-873.
SUTHERLAND D. (1974) Petrography and mineralogy of
the peralkaline silicic rocks. Bull. Volcanol. Special Vol.
38,517-547.
TAYLOR B. E., EICHELBERGER J. C. and WESTRICH
H. R. (1983) Hydrogen isotopic evidence of rhyolitic
magma degassing during shallow intrusion and eruption.
Nature 306,541-545.
THOMPSON A. O. and DODSON R. G. (1963) Geology of
the Naivasha area. Geol. Surv. Kenya Report No. 55.
THOMPSON R. N. and MACKENZIE W. S. (1967) Feldsparliquid equilibria in peralkaline acid liquids: an experimental study. Amer. J. Sci. 265,714-34.
TUTTLE O. F. and BOWEN N. L. (1958) Origin of granite
in the light of experimental
studies in the system
NaAlSi30s-KAISi30s-H20.
Geol. Soc. Amer. Mem. 74,
153 pp.
UNNI C. K. and SCHILLING J.-G. (1978) CI and Br degassing by volcanism along the Reykjanes Ridge and
Iceland. Nature 272, 19-23.
WALSH J. (1969) Mineral and thermal waters of Kenya.
XXIII International Geol. Cong. 19, 105-110.
WILCOX R. E. (1979) The liquid line of descent and variation diagrams. In The Evolution of the Igneous Rocks:
Fiftieth Anniversary Perspectives, (ed. H. S. YODER, JR.),
Chap. 7, pp. 205-232, Princeton University Press.
WILLIAMS L. A. J. (1982) Physical aspects of magmatism
in continental rifts. Continental and Oceanic Rifts Geodynamics, Series 8, 193-222.